Graphics Cards

[Tareghse]

Hey all! got me an AMD instinct mi100 (P26855-001 HP) about 2 months ago and just started to test and use it. Unfortunately, it's not detected on Ubuntu or Windows. Also it looks like the GPU is powered up and does something since see that it's getting hot. have 2 motherboards, 1. MSl x570-A pro. 2. MSI mag b560 torpedo. did enable the UEFI and "above 4G decode.." None of them detects the GPU. Or at least couldn't find it. I tried both windows and Ubuntu. On Windows, I see an undetected pcie adapter, and on Ubuntu, I see something with with 33mhz ram clock... which makes me believe my GPU is indeed working but can't figure out how to use it.

tried installing the drivers on Ubuntu, and tried is ng amdvbflash on Windows (didn't even show there).

Any help or suggestions is appreciated!

[fsadough]

MI-100 is a passive server GPU. If you install it on a non-server platform, even if you get it up and running, it will fail due to lack of proper air-flow, even if you are running it in an open system, you might have problems in future.

MI-100 is not supported under Windows. You need to install ROCm under Ubuntu for MI-100.

Make sure fTPM is enabled in system BIOS and SBIOS is up-to-date.

[Visionscaper]

@fsadough I plan to use a 6U casing with fans in front of the GPUs that will blow air in, and fans behind the GPUs that blow air out. That should work, right? Are there other specific requirements with respect to air cooling, or otherwise?

[Yehowshua]

Did you ever resolve this? I just got an Mi100 and plugged it into my Dell r720 server. Nothing shows up at all.

[Visionscaper]

@Tareghse @Yehowshua I'm also interested in this topic, just bought three AMD MI100 GPUs that came from a HP server. Did you get it to work in your machine? I plan to use a 6U casing with fans in front of the GPUs that will blow air in, and fans behind the GPUs that blow air out.

[fsadough]

I am interested to know what the end-users are planning to do with their MI-100 GPUs.

[Visionscaper]

I'm going to give ROCm a chance; going to use the MI100s for deep learning training workloads.